# Radiation Published articles for Radiation. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Analysis of LLM Vulnerability to GPU Soft Errors: An Instruction-Level Fault Injection Study DevFeed: [Analysis of LLM Vulnerability to GPU Soft Errors: An Instruction-Level Fault Injection Study]() Original publisher: [Read original article]() Author: Duo Chai, Zizhen Liu, Shuhuai Wang, Songwei Pei, Cheng Liu, Huawei Li, Shangguang Wang Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Cryptography and Security]() Topics: [Large Language Model]() Tags: [complexity](), [cs-ai](), [cs-ar](), [cs-cr](), [fault-injection](), [fault-tolerance](), [gpu](), [llm](), [llm-inference](), [model-architecture](), [parameter](), [radiation](), [reliability](), [resilience](), [satellite](), [vulnerability]() ### AI overview The paper reports an instruction-level fault-injection study of soft-error vulnerability during LLM inference, motivated by the limited analysis of modern LLMs compared with general-purpose applications and vision models. It examines how model architecture, parameter scale, and task complexity affect reliability. The abstract describes these as perspectives for understanding vulnerability and informing fault-tolerance mechanisms, but provides no specific results or comparative findings. ### Source excerpt arXiv:2601.19912v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are highly compute- and memory-intensive, posing significant demands on high-performance GPUs. At the same time, advances in GPU technology driven by shrinking transistor sizes and lower operating voltages have made these devices increasingly susceptible to soft errors, particularly in radiation-prone environments such as satellite computing scenarios. While prior work has examined GPU soft errors, most studies have focused on general-purpose applications or conventional neural networks primarily used for vision tasks such as classification and detection. However, vulnerability analysis of modern LLMs remains limited, despite their rapid adoption in diverse application scenarios. Given the unique characteristics of LLMs, their resilience to soft errors may differ substantially from earlier models. To bridge this gap, we conduct the first instruction-level fault injection study of LLM inference to analyze the factors affecting LLM vulnerability. Our approach reveals reliability characteristics from multiple perspectives, highlighting the effects of model architecture, parameter scale, and task complexity. These findings provide new insights into LLM reliability and inform the design of more effective fault tolerance mechanisms. ## Counterexample-driven autonomous research with application to excess radiated energy in a Z-pinch DevFeed: [Counterexample-driven autonomous research with application to excess radiated energy in a Z-pinch]() Original publisher: [Read original article]() Author: Bowen Zhu, Jian Wu Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [AI for science](), [agentic workflows](), [Artificial Intelligence](), [AI Research]() Tags: [agents](), [artificial-intelligence](), [arxiv](), [autonomous](), [conversion](), [cs-ai](), [diagnostics](), [energy](), [framework](), [investigation](), [language-model](), [physics](), [physics-plasm-ph](), [plasma](), [pulse](), [radiation](), [release](), [repair](), [research](), [resistive](), [retention](), [search]() ### AI overview Simjecture uses language model agents to pursue human-defined physics conjectures through counterexample-driven falsification and prospective hypothesis repair. It uses computational physics as a testbed and separates numerical instrument qualification from scientific claim review. In an 82.6-hour Z-pinch investigation, stronger coherent helical deformation increased resistive energy conversion while reducing radiation and increasing retained energy. A broader trailing-plasma profile delayed the later radiation maximum and recovered much of its early bolometric deficit; its spectral response changed during the pulse. ### Source excerpt arXiv:2611.03721v1 Announce Type: new Abstract: Recent breakthroughs in artificial-intelligence-assisted mathematics show that open-ended machine search can outperform human mathematicians, yet transferring this paradigm to physics requires refactoring the research workflow. We introduce Simjecture, an open-source framework in which language-model agents pursue human-defined physical conjectures. The framework tackles two central challenges in this transfer: first, because physics is empirical and no model is final, it structures inquiry around counterexample-driven falsification and prospective hypothesis repair; second, because physical theories must interact with the real world, it uses computational physics as an executable testbed while separating the qualification of numerical instruments from scientific claim review so solvers and diagnostics can be adapted within a recorded investigation. We apply the framework in an 82.6 h investigation to tackle the long-standing problem of radiated energy exceeding the imploding-shell kinetic energy in a Z-pinch. The investigation reveals competing effects of energy conversion, retention, and release timing. Stronger coherent helical deformation increases resistive conversion while reducing radiation and increasing retained energy. A broader trailing-plasma profile delays the later radiation maximum and recovers much of its early bolometric deficit, with a spectral response that changes during the pulse. These results provide new insight into energy conversion in high-energy-density plasma physics and demonstrate a new paradigm for physics research. ## We've received our first response from the Project Suncatcher prototype satellite. DevFeed: [We've received our first response from the Project Suncatcher prototype satellite.]() Original publisher: [Read original article]() Author: Travis Beals Published: 2026-10-30T17:00:00Z Content type: news Language: en Sources: [Google Research]() Topics: [Machine learning](), [generative-ai-tools](), [Physical AI]() Tags: [ai](), [data-center](), [gemma-4](), [google-research](), [haiku](), [machine-learning](), [milestone](), [moonshot](), [none](), [norway](), [orbit](), [payload](), [project-suncatcher](), [radiation](), [space](), [tests](), [thermal-stress](), [tpu](), [tpus]() ### AI overview Project Suncatcher's prototype satellite produced its first AI response after commissioning: during a ground-station pass, the team sent a prompt to the onboard Gemma 4 open model and received a space-themed haiku in a downlink over Svalbard. The team then tested limericks and telemetry prompts about payload status. Next, it plans to assess TPU performance under thermal stress and radiation in orbit, as a step toward twin interconnected satellites targeted for 2027. ### Source excerpt Video showing the Project Suncatcher team explaining the project and reacting to the TPU-generated haiku ## Extremely Sparse-View Computed Tomography from 2D Projections via Pose-Aware Diffusion Priors DevFeed: [Extremely Sparse-View Computed Tomography from 2D Projections via Pose-Aware Diffusion Priors]() Original publisher: [Read original article]() Author: Linrui Dai, Jinxiu Liang, Chu Zhou, Mae Yamaguchi, Ryuya Takahashi, Imari Sato Published: 2026-10-30T04:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Machine learning](), [3D]() Tags: [3d](), [artifacts](), [consistency](), [cs-ai](), [cs-cv](), [dataset](), [diffusion](), [learning](), [measurements](), [motion](), [optimization](), [projection](), [public](), [radiation](), [sampling](), [scanning](), [supervision](), [synthesis](), [train](), [training](), [volume]() ### AI overview The method targets CT reconstruction when both training and inference have only sparse projections, avoiding the need for densely scanned training volumes. It learns a pose-aware diffusion prior from one 2D projection per training sample and guides sampling with measured-view fidelity, shared-volume reprojection, and projection-physics consistency. Across two public CT datasets with 1, 4, and 8 views, it reports the strongest novel-view synthesis scores and reconstruction quality that is best or competitive against baselines trained with dense-volume supervision. ### Source excerpt arXiv:2610.38319v1 Announce Type: cross Abstract: Extremely sparse-view computed tomography (xSV-CT) is critical when scanning time, motion, or radiation tolerance limits each scan to only a few projections. In many such scenarios, the same acquisition limits also constrain dataset construction, making pre-reconstructed 3D volumes from dense scans unavailable. This exposes a gap in current sparse-view approaches: Optimization-based methods, which use the sparse measurements directly but lack data-driven priors to suppress artifacts, and learning-based methods, which use strong priors but usually require densely scanned training volumes acquired outside the xSV-CT regime. We study this train-sparse / infer-sparse setting by learning a pose-aware diffusion prior from one 2D projection per training sample, then guiding diffusion sampling with measured-view fidelity, shared-volume reprojection, and projection-physics consistency for tomographic reconstruction. On two public CT datasets across 1-, 4-, and 8-view settings, the proposed method achieves the best novel-view synthesis scores and best or competitive reconstruction quality compared with baselines that use dense-volume supervision. ## In Orbit, Useful Life Becomes a Performance Metric DevFeed: [In Orbit, Useful Life Becomes a Performance Metric]() Original publisher: [Read original article]() Author: Ron Honig Published: 2026-10-26T08:25:36Z Content type: article Language: en Sources: [EE Times]() Topics: [High-Performance Computing](), [performance-engineering](), [Benchmark]() Tags: [2026](), [about](), [ai](), [ai-hardware](), [analog](), [analysis](), [blog](), [business](), [community](), [compute](), [computing](), [cooling](), [data-centers](), [designline](), [digital](), [economics](), [education](), [ee-times](), [ee-times-university](), [eelife](), [eelive-electronic-design](), [eet](), [electrical-engineering](), [electronic-components](), [electronic-engineering-times](), [electronics](), [embedded-systems](), [fundamentals-courses](), [google](), [google-s](), [hardware](), [learning](), [life](), [mixed-signal](), [news](), [opinion](), [orbit](), [orbital-data-center](), [part-search](), [performance](), [power](), [product-search](), [project](), [project-suncatcher](), [radiation](), [satellite-communications](), [semiconductors](), [software](), [solar-power](), [space](), [tech-papers](), [technology](), [testing](), [thermal-radiation](), [useful](), [video]() ### AI overview For orbital computing, processor performance should include dependable computation delivered over the spacecraft's lifetime: a faster chip may require heavier power and cooling systems or fail sooner, reducing total useful output. The article describes major system constraints--solar generation, heat rejection in vacuum, communications, launch mass, utilization, and replacement costs. A cited one-megawatt model estimates about 5,640 m² of photovoltaic arrays and 2,500 m² of radiators under its assumptions; its cost conclusions are likewise scenario-dependent. Google's Project Suncatcher has begun testing Trillium TPUs for radiation tolerance, workload errors, and thermal behavior, while a planned 2027 optical-link experiment will examine satellite communications. These experiments are early evidence, not proof of commercial economics; sustained operation and larger systems remain untested. NASA's HPSC targets spacecraft autonomy, radiation resilience, and fault tolerance, with preliminary performance figures that use different comparisons and are not directly comparable to TPU benchmarks. ### Source excerpt Google's Project Suncatcher is testing AI hardware in space, where radiation, cooling, power, and hardware lifetime determine the economics of computing. The post In Orbit, Useful Life Becomes a Performance Metric appeared first on EE Times. ## RobustNCL: Robustifying Neuromorphic Continual Learning under Transient Faults with Lightweight Detect-and-Cure Strategy DevFeed: [RobustNCL: Robustifying Neuromorphic Continual Learning under Transient Faults with Lightweight Detect-and-Cure Strategy]() Original publisher: [Read original article]() Author: Rachmad Vidya Wicaksana Putra, Shuochen Zhou, Muhammad Shafique Published: 2026-10-26T04:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [Machine learning](), [Neural Network](), [reliability-engineering]() Tags: [cifar-10](), [continual-learning](), [cs-ai](), [cs-lg](), [cs-ne](), [datasets](), [distribution](), [dram](), [energy](), [fault-tolerance](), [hardware](), [latency](), [lightweight](), [neuromorphic](), [radiation](), [reliability](), [replacement](), [results](), [runtime](), [systems](), [techniques](), [weight]() ### AI overview RobustNCL targets transient faults in DRAM-based off-chip memory used by neuromorphic continual learning systems. It monitors weight reads for anomalous signatures and replaces affected values with zero, a minimum or maximum, or the most-probable value from the clean weight distribution; the curing values are kept in radiation-hardened memory. The abstract reports that at a high fault rate (5x10⁻⁴ for memory replay), the best variant stayed within 0.5 percentage points of fault-free accuracy and improved accuracy over an unmitigated baseline across several NCL methods and four split-image datasets, with negligible latency and energy overhead. These are abstract-level reported results; detailed evaluation conditions are not provided here. ### Source excerpt arXiv:2610.31954v1 Announce Type: cross Abstract: In neural networks, continual learning (CL) is a prominent approach for enabling deployed systems to adapt to dynamic operational environments by effectively learning from new unseen data at runtime. Recently, the neuromorphic continual learning (NCL) concept has emerged as an energy-efficient alternative to CL by leveraging sparse event-based operations. However, NCL performance can be undermined by transient faults in the hardware platforms, which may come from high-energy particle strikes at any time. Currently, NCL reliability under transient faults has not been studied, and hence the corresponding fault-tolerance techniques remain unexplored. To address this, we propose a novel RobustNCL framework to improve NCL reliability in the presence of transient faults in the DRAM-based off-chip memory of hardware platforms through a lightweight detect-and-cure strategy. Specifically, it first monitors the presence of faults during weight data access from off-chip memory, and then cures the accessed weights if anomalous signatures are detected. It provides three curing variants: replacement with zero, min/max value, and most-probable value from the original/clean weight distribution. To support this, the weight curing values are stored in the radiation-hardened memory. Experimental results show that, in the presence of high fault rates (e.g., 5x10^{-4} for memory replay), RobustNCL maintains high accuracy close to the accuracy of fault-free models (i.e., within -0.5% accuracy for the best RobustNCL variant) and significantly improves accuracy from the baseline without fault mitigation across different NCL methods and datasets (i.e., Split MNIST, Split Fashion-MNIST, Split CIFAR-10, and Split CIFAR-100), while incurring negligible latency and energy overheads. These results highlight that our RobustNCL framework is a promising approach for enabling reliable and energy-efficient NCL systems. ## Eye of Sauron: long-range hidden spy camera detection and positioning with inbuilt memory EM radiation DevFeed: [Eye of Sauron: long-range hidden spy camera detection and positioning with inbuilt memory EM radiation]() Original publisher: [Read original article]() Author: Anne Barela Published: 2026-10-25T13:30:48Z Content type: article Language: en Sources: [Adafruit Industries]() Topics: [webcam](), [communications]() Tags: [camera](), [cameras](), [detection](), [em](), [hidden](), [memory](), [positioning](), [radiation](), [range](), [video]() ### AI overview ESauron detects and locates hidden cameras by sensing electromagnetic radiation from their memory circuitry. It deliberately changes the scene in front of a camera; the resulting increase in image-encoding workload changes switching voltage regulator activity and produces a detectable EM surge at the memory clock frequency. Because it senses radiation rather than network traffic, the approach applies to wireless, wired, and offline cameras that record to a card. In tests of 50 camera products, the prototype reportedly detected all of them after four stimuli, with a range exceeding 20 meters, including through obstructions, and located each camera. These are results reported in the supplied excerpt; it does not describe test conditions or limitations. ### Source excerpt Researchers from Hunan University, UC San Diego, and Michigan State University presented ESauron at USENIX Security '24, a system that finds hidden spy cameras by listening for the electromagnetic radiation their memory gives off. The idea rests on something every camera has to do. Raw images get encoded and compressed before they are sent anywhere [...] ## This Newly Proposed Linux Filesystem Can Fix Corrupted Data DevFeed: [This Newly Proposed Linux Filesystem Can Fix Corrupted Data]() Original publisher: [Read original article]() Author: Sourav Rudra Published: 2026-10-22T14:39:35Z Content type: news Language: en Sources: [It's FOSS]() Topics: [Linux](), [btrfs](), [data](), [backups](), [Ext4](), [Tool]() Tags: [anthropic](), [btrfs](), [claude](), [error-correcting-codes](), [ext4](), [filesystem](), [fsck](), [kernel](), [linux](), [news](), [radiation](), [repair](), [satellites]() ### AI overview BeamFS proposes correcting small data errors without a second copy: Reed-Solomon codes are stored in each block, checked on every read, and used to repair damage before data reaches an application; damage beyond repair produces a read error. This may suit single-drive systems exposed to radiation, though real radiation testing has not been done. Simulated bit-flip tests found ext3 and ext4 returned wrong data without errors, Btrfs refused a read, and BeamFS returned the correct file except when a flip affected a data pointer; the tests used unequal numbers of flips and do not rank the filesystems. The 18-patch proposal is an RFC, not a merge request. BeamFS has no journal and relies on fsck after an unclean shutdown, and its README warns of data loss, security holes, and kernel crashes, advising against using it for important data. ### Source excerpt Aimed at hardware exposed to radiation, like satellites, BeamFS can repair data without keeping a second copy. ## Spatiotemporal Reconstruction of Global Radiation from Sparse Sensor Sequences DevFeed: [Spatiotemporal Reconstruction of Global Radiation from Sparse Sensor Sequences]() Original publisher: [Read original article]() Author: Kazuma Kobayashi, Tapas Tripura, Jay Phil Yoo, Samrendra Roy, Diab Abueidda, Seid Koric, Souvik Chakraborty, Syed Bahauddin Alam Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [Neural Network]() Tags: [2](), [abstract](), [arxiv](), [based](), [calendar](), [can](), [cpu](), [cs-ai](), [cs-lg](), [distributed](), [eess-sp](), [effective](), [error](), [field](), [fields](), [general](), [global](), [gpu](), [gpu-inference](), [here](), [radiation](), [sensor](), [sequences]() ### AI overview TRON reconstructs spatial physical fields from sparse histories of indirect sensors. For cosmic radiation, it estimates sea-level effective dose at 65,341 locations using data from 12 neutron monitors, with 0.096% mean relative L2 error against EXPACS-generated fields in a chronologically held-out year. Batched GPU inference takes about 1 ms per global field, compared with about 57.6 seconds for EXPACS on one CPU thread. On measured weather networks, it reconstructs temperatures at held-out stations from thermometer or barometer histories and outperforms the reported interpolation, calendar-based, and hypsometric baselines. The supplied text describes abstract-level results and does not provide further methodological detail. ### Source excerpt arXiv:2506.12045v3 Announce Type: replace Abstract: Much of the physical world is inaccessible to direct measurement and can be reached only through the sparse traces it leaves in other observable variables. Here we introduce the Temporal Radiation Operator Network (TRON), a spatiotemporal neural operator that learns proxy-to-field mappings, reconstructing otherwise unobserved physical fields from sparse, indirect sensor histories. For cosmic-radiation mapping, TRON reconstructs sea-level effective dose at 65,341 locations from histories of 12 neutron monitors, achieving 0.096\% mean relative $L_2$ error against EXPACS-generated reference fields in a chronologically held-out year; batched GPU inference evaluates a global field in approximately $\sim$1~ms, compared with $\sim$57.6~s for EXPACS on one CPU thread. On measured weather networks, TRON reconstructs temperature at held-out stations from thermometer or barometer histories, outperforming interpolation, calendar-based and hypsometric baselines. These results establish proxy-to-field reconstruction as a general route to recovering spatially distributed physical information beyond the direct reach of sparse sensor networks. ## STAR-XPC: Set Transformer for Estimating X-ray Transmission in Polymer Composites DevFeed: [STAR-XPC: Set Transformer for Estimating X-ray Transmission in Polymer Composites]() Original publisher: [Read original article]() Author: Can Erer, H. Birkan Yilmaz, Nebahat Aral Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Artificial Intelligence](), [swift-transformers]() Tags: [beam](), [chemistry](), [cross](), [cs-ai](), [cs-lg](), [dataset](), [inference](), [monte-carlo](), [neural-network](), [orders](), [pipeline](), [radiation](), [ratio](), [set](), [simulation](), [transformer](), [transmission]() ### AI overview STAR-XPC estimates X-ray transmission through polymer shielding composites in milliseconds, providing a fast surrogate for Monte Carlo simulations that can take several hours. It uses a permutation-invariant Set Transformer to process mixtures of up to 11 molecular constituents, their mass fractions, and four bulk physical descriptors; ChemBERTa supplies constituent embeddings. The model was trained on 4,697 configurations generated with PENELOPE/penEasy for polyethylene-based slabs 0.15-2.5 mm thick under 30, 80, 100, and 120 kV sources. In jointly stratified, leakage-free 10-fold cross-validation, it achieved R² = 0.9988, median absolute percentage error of 1.15%, and an alpha-trimmed error of 1.66%. These results are reported for the simulated configurations and described setup. ### Source excerpt arXiv:2610.28709v1 Announce Type: new Abstract: Evaluating the energy transmission ratio of an X-ray beam through a polymer composite is a challenging task in radiation shielding studies. The reference solution with Monte Carlo transport simulation is highly accurate but computationally expensive: each of our reference simulations is capped at $4\times10^9$ photon histories and a 20,000 s (over 5 hours) time limit, while closed-form analytical methods fail to represent such complex scenarios. We present an artificial neural network model that maps the unordered multi-set of molecular constituents, their mass fractions, and four bulk physical descriptors directly to the transmission ratio $I(t)/I_0$ in 1-3 milliseconds. The model is a Set Transformer, which makes it permutation invariant by construction and accepts mixtures of different sizes up to 11 constituents within a fixed-size padded set. Each constituent is embedded with a chemistry-pretrained transformer language model, ChemBERTa. In our proposed method, training data are produced by an automated PENELOPE/penEasy pipeline using the radiation source at 30, 80, 100, and 120 kV tube voltages over polyethylene-based shielding composite slabs of thicknesses spanning 0.15-2.5 mm. After preprocessing, the dataset contains $n = 4697$ simulated configurations on which a leakage-free, jointly stratified 10-fold cross-validation is performed, attaining mean squared error MSE = $(1.12 \pm 0.56)\times10^{-4}$, normalized-MSE percentage NMSE% = 0.0294% $\pm$ 0.0147%, and $R^2 = 0.9988$. The corresponding per-sample percentage errors give an alpha-trimmed ($\alpha = 0.10$) absolute percentage error (APE) of 1.66% and a median APE of 1.15%. Our model's inference is several orders of magnitude faster than the underlying Monte Carlo simulation (sub-millisecond per query vs. several hours for the reference run) with effective accuracy in predicting $I(t)/I_0$. ## Injection of Space Radiation Errors on Low-power NVM using Generative Deep Learning DevFeed: [Injection of Space Radiation Errors on Low-power NVM using Generative Deep Learning]() Original publisher: [Read original article]() Author: Eleonora Giudici, Antonio Miele, Luca Mottola Published: 2026-10-21T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Deep learning](), [Machine learning](), [Artificial Intelligence]() Tags: [cs-ar](), [cs-lg](), [cubesat](), [deep-learning](), [errors](), [generative](), [low-earth-orbit](), [low-power](), [memory](), [model](), [non-volatile-memory](), [nvm](), [pattern](), [patterns](), [radiation](), [space]() ### AI overview The authors train a generative deep learning model to reproduce spatial patterns of bit errors observed in an off-the-shelf low-power FRAM chip on a CubeSat in Low Earth Orbit. They identify errors by comparing repeated read/write-test snapshots with a reference bitmap, then use the observed patterns to generate synthetic errors for application-level experiments. Preliminary validation reports that 92% of generated bit flips correspond to observed flips and that the model captures 91% of observed flips; across 10,000 generated patterns, the Wasserstein distance between spatial distributions is 0.073. These are preliminary results based on the described satellite data. ### Source excerpt arXiv:2610.26724v1 Announce Type: cross Abstract: We develop an error model for a Non-Volatile Memory (NVM) device that represents the effects of physical faults experienced in a CubeSat mission in Low Earth Orbit (LEO). The CubeSat carries an off-the-shelf low-power Ferroelectric RAM (FRAM) chip and runs repeated read/write tests according to a predefined pattern. We compare every memory snapshot received from the satellite with a reference bitmap to identify bit errors. We visually analyze erroneous values distributions in the memory snapshots and we understand that they are frequently distributed according to regular spatial patterns. We define a generative deep learning technique to automatically learn these spatial patterns and to synthetically generate additional error patterns mimicking the observed ones. The model is capable at producing memory errors for general application-level experimentation. Our preliminary validation indicates that when comparing the model output with the data we collect, 0.92 of the generated bit flips correspond to bit flips in the memory snapshots and the model captures 0.91 of the observed bit flips. A comparison of the spatial distribution of bit flips in 10,000 generated error patterns also provides a Wasserstein distance of 0.073, indicating that the two distributions differ by roughly 7% of the maximum possible displacement of their probability masses. ## A Radiation-Hardened, Fault-Aware Quantized Neural Network That Self-Heals DevFeed: [A Radiation-Hardened, Fault-Aware Quantized Neural Network That Self-Heals]() Original publisher: [Read original article]() Author: Peter Xiangyuan Ma, Joshua S. Bloom Published: 2026-10-21T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [quantization](), [Neural Network](), [Deep learning]() Tags: [alignment](), [astro-ph-im](), [benchmarks](), [cs-lg](), [data-driven](), [ecc](), [fault-tolerance](), [hardware](), [hep-ex](), [inference](), [neural](), [neural-network](), [nucl-ex](), [physics-ins-det](), [radiation](), [rank](), [redundancy](), [regression](), [runtime]() ### AI overview Projection Healing repairs flagged bit-flipped weights in quantized neural networks at runtime on commodity hardware, without storing extra replicas. It extends Fault-Aware Training with an alignment penalty that concentrates undamaged activations and weights in a low-rank subspace, then reconstructs corrupted coordinates from uncorrupted ones. On six regression benchmarks, enabling runtime healing on a fixed FAT checkpoint reduced native-unit excess integral by about 23%. The method extended median operational scrub intervals by 2.04x at +5% RMSE degradation and 1.96x at +50%; classical SECDED achieved 1.62x and 1.07x, respectively. These results position the method as a complement to hardware redundancy, ECC, and FAT techniques. ### Source excerpt arXiv:2610.27621v1 Announce Type: cross Abstract: Deep learning (DL) models are increasingly being deployed into radiation-challenging environments in spacecraft, particle accelerators, and, soon, space-based data centers. Instantiated as a learned orchestration of numbers stored in persistent memory, models are susceptible to continual degradation as high-energy particles corrupt data. At inference time, new data passed through the network fundamentally break the off-line assumptions and guarantees of a static model determined during training. To preserve the quality and efficacy of the models, fault tolerance against bit flips is a clear objective. In this work we developed a runtime method for repairing the quantized weights of neural networks (NNs) after bit flips on commodity, commercial off-the-shelf (COTS) hardware, without storing extra replicas. Our work extends the well-established Fault-Aware Training (FAT) paradigm by adding an alignment penalty to concentrate undamaged activations and weights into a low-rank subspace, and we then reconstruct flagged corrupted coordinates from the uncorrupted survivors. We call this method {\it Projection Healing} and apply it to six regression benchmarks on quantized NNs. Holding a FAT checkpoint fixed and toggling only the runtime heal gives a further ${\sim}23\%$ reduction in the native-unit excess integral. Across those datasets the method extends the operational scrub interval by a median of $2.04\times$ at a $+5\%$ RMSE degradation and $1.96\times$ at $+50\%$, compared with $1.62\times$ and $1.07\times$ for classical Single Error Correction, Double Error Detection (SECDED). This introduces a new path towards designing resilient data-driven models, complementing established hardware redundancy, ECC, and FAT-based techniques. ## Our Project Suncatcher prototype satellite is in orbit. DevFeed: [Our Project Suncatcher prototype satellite is in orbit.]() Original publisher: [Read original article]() Author: Travis Beals Published: 2026-10-01T23:30:00Z Content type: news Language: en Sources: [Google Research]() Topics: [multimodal-ai]() Tags: [google-research](), [launch](), [mission](), [none](), [paper](), [partnership](), [project-suncatcher](), [prototype](), [radiation](), [satellite](), [space](), [spacex](), [stress](), [thermal](), [transporter-18]() ### AI overview Google says its Project Suncatcher prototype satellite, built with Planet, launched into orbit on SpaceX's Transporter-18 rideshare mission and is operating as expected. The research will examine whether space could eventually support scalable machine learning infrastructure, including how Google TPUs handle radiation, thermal extremes, and physical stress during spaceflight. ### Source excerpt Video showing a rocket launch ## Google's Project Suncatcher will test whether TPUs can operate in orbit DevFeed: [Google's Project Suncatcher will test whether TPUs can operate in orbit]() Original publisher: [Read original article]() Author: Steven J. Vaughan-Nichols Published: 2026-09-28T16:22:00Z Content type: news Language: en Sources: [The New Stack]() Topics: [AI Infrastructure](), [NVIDIA Vera Rubin](), [NVIDIA DRIVE]() Tags: [ai](), [ai-data-centers](), [ai-infrastructure](), [all](), [carnegie-mellon](), [carnegie-mellon-university](), [cooling](), [elon-musk](), [hardware](), [infrastructure](), [new-york](), [new-york-times](), [nvidia](), [project-suncatcher](), [radiation](), [research](), [vera-rubin]() ### AI overview Google's Project Suncatcher will send a prototype satellite carrying four Tensor Processing Units into orbit to test whether machine-learning infrastructure can operate in space. The research will examine challenges including radiation, cooling, and launch stresses, while exploring whether solar-powered satellites connected by optical links could support large-scale workloads. ### Source excerpt While SpaceX and Nvidia plan to place a space-optimized Vera Rubin AI system in orbit by late 2027, Google's far The post Elon Musk says space will soon hold nearly all compute. Google is still finding out if its chips can work there. appeared first on The New Stack. ## Canadian Startup Sees Success in Space Qualification DevFeed: [Canadian Startup Sees Success in Space Qualification]() Original publisher: [Read original article]() Author: Gary Hilson Published: 2026-09-22T16:21:10Z Content type: news Language: en Sources: [EE Times]() Topics: [Hardware](), [systems]() Tags: [analog](), [analysis](), [business](), [community](), [designline](), [digital](), [education](), [ee-times](), [ee-times-university](), [eelife](), [eelive-electronic-design](), [eet](), [electrical-engineering](), [electronic-components](), [electronic-engineering-times](), [electronics](), [embedded-systems](), [engineering](), [fundamentals-courses](), [hardware](), [learning](), [military-and-aerospace](), [mixed-signal](), [new-space-electronics](), [news](), [opinion](), [part-search](), [product-search](), [radiation](), [semiconductors](), [software](), [space-grade-reliability](), [startup](), [tech-papers](), [technology](), [testing](), [video]() ### AI overview Novigrad, a Saskatoon-based startup, provides radiation testing and qualification for electronics intended for space and other harsh environments. The company addresses demand from Canadian space and semiconductor organizations that previously relied on facilities outside Canada. ### Source excerpt Novigrad is forging a pathway for faster, more accessible radiation testing as a growing wave of space missions drives demand for reliable, resilient electronics. The post Canadian Startup Sees Success in Space Qualification appeared first on EE Times.